LaunchDarkly CTO Cameron Etezadi discusses the shift from large engineering teams to small AI-augmented ones, introducing the 'dark factory' concept for safely observing and deploying AI-generated code using runtime agent frameworks. The newsletter also covers: new research on CO2 levels impairing developer cognition, the GLM-5.2 open source model's impact on AI cost pressure and model routing, the messy reality of running AI locally for coding (RAM constraints, reasoning loops, context window blowouts), the decline of deep reading accelerated by AI-generated content, and a story about shipping a Facebook feature so fast it triggered an executive emergency meeting — used as a cautionary tale about autonomous agents running without visibility.

5m read timeFrom devinterrupted.substack.com
Post cover image
Table of contents
1. Are high CO2 levels killing your team’s cognition?2. Why GLM-5.2 is forcing a rethink on AI costs3. The messy reality of running AI locally4. The death of deep reading5. Life beyond tokenmaxxing6. When shipping too fast panics the C-Suite
117 Impressions